A cloud phone authentication method and related equipment
By acquiring user behavior data at the cloud phone hardware layer, extracting network and threat intelligence using time-series features to calculate risk levels, dynamically selecting multi-factor authentication strategies, and generating dynamic tokens bound to the physical characteristics of the device, the security deficiencies of static passwords and single biometric features in cloud phone authentication schemes are resolved, achieving efficient user identity authentication and continuous authorization.
Patent Information
- Application Number
- CN202510997009.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-18
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2045-07-18
AI Technical Summary
Existing cloud phone authentication solutions rely on static passwords or single biometric features, which cannot dynamically detect abnormal user behavior, leading to security failures when devices are hijacked or biometric data is leaked.
By acquiring user behavior data from the cloud phone hardware layer, extracting network-generated user behavior deviation probabilities using time-series features, calculating risk levels by combining threat intelligence weights, dynamically selecting multi-factor authentication strategies, and generating dynamic tokens bound to the physical characteristics of the device for authentication.
It enables continuous authentication and dynamic authorization of user identity in a cloud phone environment, effectively responding to device hijacking and spoofing attacks, and improving the reliability and adaptability of authentication.
Smart Images

Figure CN120768632B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of network security technology, and in particular to an authentication method and related equipment for cloud phones. Background Technology
[0002] With the rapid development of mobile internet technology, cloud phones, as an emerging computing form, have achieved lightweight terminal devices by cloudifying computing resources. However, their reliance on network communication has also introduced new security risks. Existing authentication schemes mostly rely on static passwords or single biometric features (such as fingerprints or facial recognition), which have inherent flaws in dealing with device hijacking and biometric replay attacks. Once biometric data is leaked, it will lead to permanent security failure, and static authentication cannot dynamically detect abnormal user behavior. Therefore, there is an urgent need for an authentication method for cloud phones to solve the aforementioned technical problems. Summary of the Invention
[0003] The summary section introduces a series of simplified concepts, which will be further explained in detail in the detailed description section. This summary section is not intended to limit the key and essential technical features of the claimed technical solutions, nor is it intended to define the scope of protection of the claimed technical solutions.
[0004] Firstly, this application provides a cloud phone authentication method, including:
[0005] Acquire user behavior data at the cloud phone hardware layer, including user interaction characteristics, device fingerprints, network environment data, and sensor data;
[0006] Based on user behavior data, user behavior is modeled through a time-series feature extraction network to generate user behavior deviation probability.
[0007] Risk level scores are calculated based on the probability of behavioral deviation and threat intelligence weights.
[0008] Based on the risk level score, a preset multi-factor authentication strategy is selected;
[0009] Based on a preset multi-factor authentication strategy, a dynamic token bound to the physical characteristics of the device is generated;
[0010] Based on dynamic tokens, user identity is authenticated to determine the user's identity verification result.
[0011] In some implementations, acquiring user behavior data at the cloud phone hardware layer includes:
[0012] Based on probes deployed in the Hypervisor layer, raw signal data from physical sensor registers is read;
[0013] Based on the gyroscope and accelerometer sample values in the original signal data, a three-dimensional motion trajectory model of the sensor data is generated through a Kalman filter.
[0014] Based on a preset anti-collision algorithm, a device fingerprint is generated by processing the device's unique identifier set, which includes IMEI, MAC address, and SIM card ICCID.
[0015] Based on the packet header information captured by the network interface, network environment data is generated by parsing. The network environment data includes the rate of change of IP geographical location and the base station handover frequency.
[0016] Based on the user input event stream, click coordinate distribution and swipe trajectory patterns are extracted to generate user interaction features;
[0017] The three-dimensional motion trajectory model, device fingerprint, network environment data, and user interaction features are integrated into user behavior data.
[0018] In some implementations, user behavior is modeled using a time-series feature extraction network based on user behavior data to generate user behavior deviation probabilities, including:
[0019] Based on the click coordinate distribution and swipe trajectory pattern in user interaction features, an operation event sequence is generated;
[0020] Based on the sequence of operation events, behavioral feature vectors are determined through a pre-set temporal feature extraction network.
[0021] Based on behavioral feature vectors and a preset user behavior baseline model, the degree of behavioral deviation is determined by cosine similarity calculation.
[0022] Based on the mapping relationship between behavioral deviation and preset deviation threshold, the probability of user behavior deviation is determined.
[0023] In some implementations, a risk level score is calculated based on the probability of behavioral deviation and threat intelligence weights, including:
[0024] The first rating weight coefficient is determined based on the probability of user behavior deviation;
[0025] The second scoring weight coefficient is determined based on real-time risk indicators from external threat intelligence sources.
[0026] An initial risk score is generated by multiplying the first score weight coefficient by the probability of user behavior deviation, and then adding the product of the second score weight coefficient and the threat intelligence weight.
[0027] The initial risk score is normalized based on the preset risk mapping rule to determine the risk level score. The preset risk mapping rule is the correspondence between the score range and the risk level.
[0028] In some implementations, a preset multi-factor authentication strategy is selected based on the risk level score, including:
[0029] Based on the risk level score and the first preset threshold, the first risk level range is determined;
[0030] The second risk level range is determined based on the risk level score and the second preset threshold.
[0031] When the risk level score is in the first risk level range, the target authentication strategy is determined to be the first type of authentication strategy, which combines static passwords and dynamic tokens.
[0032] When the risk level score is in the second risk level range, the target authentication strategy is determined to be the second type of authentication strategy that adds device fingerprint verification.
[0033] When the risk level score is less than the second preset threshold, the target authentication strategy is determined to be the third type of authentication strategy that combines voiceprint recognition and geofence verification.
[0034] In some implementations, a dynamic token bound to the physical characteristics of the device is generated based on a preset multi-factor authentication strategy, including:
[0035] Based on the target type of the preset multi-factor authentication strategy, determine the dynamic token algorithm type;
[0036] The device battery temperature value based on sensor data is used to generate a first salt value through a first preset scaling transformation.
[0037] The second salt value is generated based on the screen brightness value from sensor data through an 8-bit unsigned integer conversion.
[0038] An initial token sequence is generated based on the true random number seed output by the quantum random number generator;
[0039] Based on the first salt value and the second salt value, a hash binding operation is performed on the initial token sequence to obtain the hash binding operation result;
[0040] The current operation sequence identifier is determined based on the globally monotonically increasing sequence number rule;
[0041] A dynamic token is generated based on the hash binding operation result and the current operation sequence identifier to bind the physical characteristics of the device.
[0042] In some implementations, user authentication is performed based on a dynamic token to determine the user authentication result, including:
[0043] The sequence continuity verification result is determined based on the current operation sequence identifier in the dynamic token and the server-side globally monotonically increasing sequence number.
[0044] The token hash reconstruction value is determined based on the device battery temperature value, screen brightness value, and quantum random number generator.
[0045] The hash consistency verification result is determined based on the hash binding operation result of the token hash reconstruction value and the dynamic token;
[0046] The user authentication result is determined by performing a logical AND operation between the sequence continuity verification result and the hash consistency verification result.
[0047] Secondly, this application proposes an authentication device for a cloud phone, comprising:
[0048] The behavior data acquisition unit is used to acquire user behavior data at the cloud phone hardware layer. The user behavior data includes user interaction characteristics, device fingerprints, network environment data, and sensor data.
[0049] The deviation probability generation unit, based on user behavior data, models user behavior through a time-series feature extraction network to generate user behavior deviation probabilities.
[0050] The risk level scoring unit calculates the risk level score based on the probability of behavioral deviation and the weight of threat intelligence.
[0051] The authentication strategy selection unit selects a preset multi-factor authentication strategy based on the risk level score;
[0052] The dynamic token generation unit generates a dynamic token bound to the physical characteristics of the device based on a preset multi-factor authentication strategy.
[0053] The identity verification unit performs authentication operations on the user's identity based on the dynamic token and determines the user's identity verification result.
[0054] Thirdly, an electronic device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program stored in the memory to implement the authentication method of the cloud phone according to any one of the first aspects.
[0055] Fourthly, this application proposes a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the authentication method for a cloud phone according to any one of the first aspects.
[0056] In summary, this application determines whether a device has been hijacked by comprehensively collecting and modeling user interaction, device, network, and sensor data; it enhances token security by dynamically switching algorithms and introducing quantum random numbers based on behavioral analysis results; it improves authentication flexibility by dynamically adapting multi-factor strategies based on risk levels and supporting gray-scale upgrades; it quickly isolates abnormal operations through lightweight containerization and micro-segmentation technology; and it achieves continuous authentication and dynamic authorization for cloud phone users, effectively addressing security threats such as device hijacking and spoofing attacks. While ensuring security, it adapts to the limited hardware resources of cloud phones, improving the reliability and adaptability of cloud phone identity authentication.
[0057] The cloud phone authentication method proposed in this application, along with other advantages, objectives, and features of this application, will be partly apparent from the following description and partly understood by those skilled in the art through research and practice of this application. Attached Figure Description
[0058] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit this specification. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:
[0059] Figure 1 A schematic diagram of an authentication method for a cloud phone provided in this application embodiment;
[0060] Figure 2 A schematic diagram of the authentication device structure for a cloud phone provided in an embodiment of this application;
[0061] Figure 3 This is a schematic diagram of the structure of an authentication electronic device for a cloud phone, provided as an embodiment of this application. Detailed Implementation
[0062] The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus. The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them.
[0063] Please see Figure 1 This is a schematic diagram of a cloud phone authentication method provided in an embodiment of this application, which may specifically include:
[0064] S110. Obtain user behavior data from the cloud phone hardware layer, including user interaction characteristics, device fingerprint, network environment data, and sensor data.
[0065] For example, this step directly acquires raw signals from hardware sensors and network interfaces by breaking through the low-level data acquisition mechanism of traditional operating system permission restrictions. Specifically, it utilizes a lightweight probe (e.g., developed based on the KVM virtio driver) deployed with Hypervisor layer privileges to read physical sensor registers and network packet header information in real time through memory mapping technology. This direct hardware-level acquisition method avoids the potential tampering risks of the Guest OS, ensuring the originality and integrity of user interaction features, device fingerprints, network environment data, and sensor data. User interaction features are represented by extracting click coordinate distribution and swipe trajectory patterns from the input event stream to quantify operating habits; device fingerprints are generated by fusing unique device identifiers such as IMEI and MAC address based on an anti-collision algorithm to produce an irreversible device identifier; network environment data involves parsing packet header information and dynamically calculating IP mutation rate and base station handover frequency; sensor data is generated by fusing raw sampling values from gyroscopes and accelerometers to construct a three-dimensional motion trajectory model of the device.
[0066] The collected multi-dimensional data, after initial integration, forms a user behavior dataset, providing an immutable input source for subsequent behavior modeling. It should be noted that this process only completes the acquisition and basic encapsulation of raw signals, without involving complex calculations (such as behavior analysis or risk assessment). By controlling the data acquisition latency to within 2ms, it adapts to the resource-constrained characteristics of cloud phones, providing underlying data support for real-time authentication decisions.
[0067] S120. Based on user behavior data, model user behavior through a time-series feature extraction network to generate user behavior deviation probability;
[0068] For example, based on the multi-dimensional user behavior data (including interaction features, device fingerprints, etc.) obtained in step S110, this step constructs a dynamic behavior model through a temporal feature extraction network. First, user interaction features (such as click coordinate distribution and swipe trajectory patterns) are transformed into a sequence of continuous operation events. Then, LSTM or a temporal convolutional network (TCN) is used to extract deep features from the sequence, generating a behavioral feature vector representing the current operating habits. In terms of network design, a 128-unit LSTM hidden layer is preferentially used (which can be reduced to 64 units when resources are limited), and the input length is optimized to 50-70 consecutive operation events to ensure efficient capture of temporal dependencies in cloud phones with limited computing power.
[0069] By calculating cosine similarity, the deviation between the current behavior feature vector and the preset user behavior baseline model is compared in real time. Based on a federated learning framework, normal operating patterns of multiple users are aggregated, and Laplace noise is injected during local model updates to meet differential privacy requirements. When the cosine similarity is lower than a preset deviation threshold (i.e., a significant difference between historical patterns and current operations), an alert is triggered and a high deviation probability value is output. This probability value directly reflects the risk intensity of device hijacking or spoofing. This process only outputs a probability scalar and does not involve specific authentication decisions, providing normalized input for subsequent risk scoring.
[0070] S130. Calculate the risk level score based on the probability of behavioral deviation and the weight of threat intelligence.
[0071] For example, this step constructs a quantitative risk assessment model by dynamically fusing internal behavioral anomalies with external threat profiles. The user behavior deviation probability serves as an input parameter, directly reflecting the temporal feature extraction network's judgment of the legitimacy of the current operation (e.g., 0.8 indicates an 80% probability of anomaly). This probability value is generated by mapping the cosine similarity calculation in step S120 to a preset deviation threshold, and is essentially a risk estimate of device hijacking or spoofing behavior.
[0072] Real-time risk indicators from external threat intelligence sources (such as high-risk IP databases or attack signature databases) are introduced and quantified into threat intelligence weights. An initial risk score is generated by linearly superimposing the probability of behavioral deviation with the threat intelligence weights using a preset scoring formula. This formula employs a weighted summation mechanism (e.g., initial risk score = k1 × probability of behavioral deviation + k2 × threat intelligence weight), where the weight coefficients are dynamically configured according to the cloud phone security policy to ensure a coordinated assessment of internal behavioral anomalies and the external threat environment.
[0073] S140. Based on the risk level score, select a preset multi-factor authentication strategy;
[0074] For example, this step dynamically adapts the authentication strategy by setting a preset risk threshold range. First, security levels are divided based on a quantified risk score. When the score is in the high-trust range (e.g., ≥ the first threshold), a basic authentication combination (static password + dynamic token) is used. When the risk is medium (between the first and second thresholds), device fingerprint verification is added to enhance security. When high risk is detected (e.g., < the second threshold), voiceprint recognition and geofencing dual verification are enabled. This tiered mechanism essentially ensures user experience in low-risk scenarios and enhances defense strength in high-risk scenarios by dynamically switching the number and type of authentication factors.
[0075] The policy switching process incorporates a canary rollout capability to address persistent threats. Upon initial detection of an anomaly, only the session duration is extended (e.g., from 5 minutes to 2 hours) to avoid frequent authentication interference. If multiple anomalies occur consecutively (e.g., 3 times), a forced upgrade to biometric authentication is implemented. This mechanism balances security and operational smoothness through a tiered response, preventing performance fluctuations in resource-constrained cloud phones due to sudden changes in security policies. The policy library pre-configures multiple authentication combination logics to ensure real-time matching between risk levels and authentication strength.
[0076] S150. Based on a preset multi-factor authentication strategy, generate a dynamic token that is bound to the physical characteristics of the device.
[0077] For example, this step achieves deep binding between authentication strength and device status through a dynamic token generation mechanism. Based on the target type of the selected authentication strategy, the token generation algorithm is adaptively switched: when the behavior score is normal, a time-window-constrained TOTP algorithm is used to ensure basic security; when anomalies such as IP mutations or high-frequency clicks are detected, the algorithm switches to an event-driven HOTP algorithm and overlays the behavior feature hash value to achieve risk-responsive token updates.
[0078] Furthermore, a quantum random number generator provides truly random seeds with an entropy value ≥0.98, enhancing the token's resistance to brute-force attacks. Simultaneously, each token is bound to a globally monotonically increasing sequence number, with the server verifying sequence continuity to defend against replay attacks. This process ensures token randomness while establishing a device-to-token correspondence through physical feature binding and a sequence number mechanism, preventing cross-device reuse and man-in-the-middle attacks at the source.
[0079] S160. Based on the dynamic token, perform authentication operations on the user's identity and determine the user's authentication result.
[0080] For example, this step implements closed-loop authentication of user identity through a multi-factor authentication mechanism. Sequence continuity verification is based on the current operation sequence identifier embedded in the dynamic token, which is compared in real-time with a globally monotonically increasing sequence number maintained by the server to verify the continuity of the operation sequence (allowing a preset tolerance window). If the sequence number is not continuously increasing (e.g., missing or duplicated), it is determined that there is a risk of replay attack or man-in-the-middle hijacking. Device status binding verification utilizes real-time collected device battery temperature values, screen brightness values, and the output of a quantum random number generator to reconstruct the token hash value; by comparing the reconstructed value with the original hash binding operation result in the dynamic token, the integrity of the device's physical state is verified. If the hash values are inconsistent, it indicates an abnormal device environment (e.g., simulator tampering or sensor deception).
[0081] The authentication result is output by performing a logical AND operation based on the above dual verification results (only if both pass). Successful authentication is indicated by a continuous sequence and consistent device status, authorizing the user to access cloud phone resources; failed authentication triggers a virtual environment isolation operation, blocking the attack link and initiating the evidence collection process. This mechanism ensures identity legitimacy from different dimensions such as operation sequence and device physical status, realizing a dynamic authorization closed loop under a zero-trust architecture.
[0082] In summary, this application's embodiments overcome the permission restrictions of traditional operating systems by directly deploying probes at the cloud phone's Hypervisor layer to capture raw hardware-level signals (including gyroscope attitude, accelerometer sampling values, and network data packet headers), ensuring the immutability of user interaction features, device fingerprints, network environment data, and sensor data. A temporal feature extraction network is used to dynamically model continuous operation event sequences. Cosine similarity metric is used to quantitatively calculate the real-time deviation probability from the user behavior baseline model, and combined with external threat intelligence weights to generate accurate risk scores. Multi-factor authentication strategies are dynamically switched based on risk level scores (e.g., 1FA for low risk, 3FA for high risk), achieving flexible matching between authentication strength and security threats. A quantum random number generator generates true random seeds, and device battery temperature and screen brightness are linearly transformed and used as salt values in token hash calculations, deeply binding dynamic tokens to the device's physical state and blocking cross-device reuse risks from the source. When abnormal operations are detected, a closed-loop verification mechanism using sequence continuity verification and token hash reconstruction values is used to determine identity legitimacy, authorizing access only when verification is successful. Ultimately, a proactive defense loop is achieved in the resource-constrained cloud phone environment, reducing the false negative rate of spoofing attacks. At the same time, a balance between security and system performance is achieved through lightweight container isolation and canary upgrade mechanisms.
[0083] In some instances, user behavior data at the cloud phone hardware layer is obtained, including:
[0084] Based on probes deployed in the Hypervisor layer, raw signal data from physical sensor registers is read;
[0085] Based on the gyroscope and accelerometer sample values in the original signal data, a three-dimensional motion trajectory model of the sensor data is generated through a Kalman filter.
[0086] Based on a preset anti-collision algorithm, a device fingerprint is generated by processing the device's unique identifier set, which includes IMEI, MAC address, and SIM card ICCID.
[0087] Based on the packet header information captured by the network interface, network environment data is generated by parsing. The network environment data includes the rate of change of IP geographical location and the base station handover frequency.
[0088] Based on the user input event stream, click coordinate distribution and swipe trajectory patterns are extracted to generate user interaction features;
[0089] The three-dimensional motion trajectory model, device fingerprint, network environment data, and user interaction features are integrated into user behavior data.
[0090] For example, this step directly accesses the physical hardware layer through a lightweight probe deployed at the Hypervisor layer, overcoming the permission restrictions of traditional Guest operating systems. The probe is developed based on the KVM virtio driver and uses memory mapping technology to read raw signal data from sensor registers, achieving hardware-level data acquisition. Specifically, the raw sample values from the gyroscope and accelerometer are captured in real-time at a 100Hz sampling rate, ensuring millisecond-level timeliness of motion data. This mechanism ensures that the acquired raw signals are not tampered with at the operating system layer, guaranteeing the authenticity and integrity of user behavior data from the data source. The acquisition process strictly controls latency to within 2ms, adapting to the resource-constrained characteristics of cloud phones.
[0091] For motion sensor data, gyroscope attitude angles and accelerometer 3D axial data are dynamically fused using a Kalman filter. The filter constructs a state transition matrix based on kinematic equations and iteratively optimizes the trajectory coordinates through a prediction-correction mechanism to generate a 3D motion trajectory model of the device. This model can detect abnormal jitter patterns (such as linear trajectory deviations in robotic arm operations), with the gyroscope angular velocity variance threshold set at 0.05 rad. 2 / s2, the dynamic range of the accelerometer is limited to ±8g to ensure that the trajectory accuracy meets the requirements of safety analysis.
[0092] Device fingerprint generation employs a pre-defined anti-collision algorithm to process the set of unique device identifiers. Specifically, the IMEI, MAC address, and SIM card ICCID are extracted using the fuzzy hash algorithm TLSH. The identifiers are first converted into fixed-length byte streams, then divided into blocks using a sliding window, and the hash digests of each block are calculated. Finally, these are fused to generate a 512-bit anti-collision fingerprint. This process ensures that the hash Hamming distance between different devices is greater than a pre-defined security threshold (typically ≥120 bits), effectively preventing device forgery attacks.
[0093] Network environment data parsing is based on packet header information captured from network interfaces. By parsing the source address field in the IP packet header and combining it with a geographic location database, the IP mutation rate is calculated in real time (e.g., displacement distance per unit time divided by the time interval; a speed > 800 km / h is marked as abnormal). Simultaneously, the frequency of cell identifier changes in base station handover requests is statistically analyzed to generate a base station handover frequency index. The packet sampling period is set to 5 times per second to balance real-time detection with resource overhead.
[0094] User interaction features are extracted from the input event stream. For touch event sequences, the screen coordinates (x, y) of each click are recorded and a two-dimensional Gaussian distribution model is constructed to generate click coordinate distribution features. For swipe operations, the coordinates of the starting point, ending point, and movement direction angle are extracted, and swipe trajectory pattern features are generated by fitting a Bézier curve. The event stream is sorted by timestamp, with a window size of 50 milliseconds to ensure the temporal integrity of the operation sequence.
[0095] Finally, the 3D motion trajectory model, device fingerprint, network environment data, and user interaction features are structurally integrated. Data packets are encapsulated in JSON format, containing independent fields for four data types and a unified timestamp. After encryption, the data packets are uploaded to the edge analytics node via a TLS 1.3 channel, with transmission latency controlled within 2 milliseconds, meeting the timeliness requirements of real-time authentication for cloud phones.
[0096] In summary, this application's embodiments ensure data immutability through direct acquisition at the Hypervisor layer; the 3D trajectory model constructed using Kalman filtering accurately identifies simulated robotic arm operations; the 512-bit device fingerprint generated by the TLSH algorithm effectively defends against collision attacks; IP mutation rate and base station switching frequency quantify network anomalies; click distribution and sliding trajectory modeling accurately capture operational features; and structured data encapsulation and encrypted transmission balance efficiency and security. Ultimately, this achieves highly reliable acquisition of multi-dimensional behavioral data under hardware resource-constrained environments, providing a solid foundation for dynamic risk authentication.
[0097] In some instances, user behavior is modeled using time-series feature extraction networks based on user behavior data, generating user behavior deviation probabilities, including:
[0098] Based on the click coordinate distribution and swipe trajectory pattern in user interaction features, an operation event sequence is generated;
[0099] Based on the sequence of operation events, behavioral feature vectors are determined through a pre-set temporal feature extraction network.
[0100] Based on behavioral feature vectors and a preset user behavior baseline model, the degree of behavioral deviation is determined by cosine similarity calculation.
[0101] Based on the mapping relationship between behavioral deviation and preset deviation threshold, the probability of user behavior deviation is determined.
[0102] For example, based on the click coordinate distribution and swipe trajectory patterns in user interaction features, a time-continuous sequence of operation events is generated. Specifically, for click operations, a set of screen touch point coordinates is extracted and a two-dimensional probability distribution model is constructed to generate coordinate distribution features representing click position habits; for swipe operations, the starting point coordinates, ending point coordinates, and movement direction angle are analyzed, and a standardized swipe trajectory pattern is generated through Bézier curve fitting. After sorting the above features by timestamp, events are sliced at preset time windows (preferably 50 milliseconds) and integrated into a structured sequence of operation events containing event type, coordinate parameters, and time intervals. This process ensures event continuity through a sliding window mechanism, providing standardized input for temporal feature extraction.
[0103] Based on the sequence of operation events, a behavioral feature vector representing user operation habits is determined through a pre-defined temporal feature extraction network. The preferred temporal feature extraction network is a Long Short-Term Memory (LSTM) network or a Temporal Convolutional Network (TCN), with the LSTM having a preset number of hidden layer units of 128 (which can be dynamically adjusted to 64-256 depending on cloud phone computing resources). The network input length is optimized to 50-70 consecutive operation events. An embedding layer maps event types and coordinate parameters into a dense vector. Deep temporal dependency features are extracted through multi-layer nonlinear transformations of the temporal network, ultimately outputting a 512-dimensional behavioral feature vector. This vector incorporates the spatiotemporal pattern characteristics of user operations, providing a high-dimensional feature representation for calculating behavioral deviation.
[0104] Based on behavioral feature vectors and a preset user behavior baseline model, real-time behavioral deviation is determined through cosine similarity calculation. The user behavior baseline model is constructed using a federated learning framework, aggregating historical normal operation data from multiple users to train the initial model. When updating the model on the local device, Laplace noise (noise scale coefficient ε is preset to 0.3-0.7) is injected to meet differential privacy requirements. During calculation, the current behavioral feature vector and the corresponding feature vector of the baseline model are multiplied by a dot product, and then divided by the product of the magnitudes of the two vectors to generate a cosine similarity value in the range [-1, 1]. Subtracting the cosine similarity value from 1 yields the behavioral deviation (e.g., if the similarity is 0.6, the deviation is 0.4). This deviation directly reflects the degree of difference between the current operation and the historical normal pattern.
[0105] Based on the behavioral deviation degree and the output range of the preset user behavior baseline model, the user behavior deviation probability is determined. The behavioral deviation probability is directly output by the temporal feature extraction network, and its numerical range is normalized to the [0,1] interval, representing the possibility that the current operation sequence deviates from the normal pattern. Specifically, when the behavioral deviation degree is greater than the preset deviation threshold of 0.5 (corresponding to a cosine similarity of less than 0.5), the temporal network outputs a high probability value (e.g., 0.8), indicating a high risk of device hijacking or spoofing; when the behavioral deviation degree is less than or equal to 0.5, it outputs a low probability value (e.g., 0.3), indicating that the operation conforms to the expected pattern. At the same time, the user behavior baseline model is updated every 30 seconds through a federated learning mechanism to ensure that the deviation probability calculation is dynamically optimized as the user behavior pattern evolves.
[0106] In summary, this application's embodiments overcome the limitations of traditional static behavior models by standardizing the encapsulation of operation event sequences, designing a lightweight temporal feature extraction network, calculating the quantized deviation of cosine similarity, and employing a dynamic threshold mapping mechanism. Through temporal modeling of continuous operation events, it accurately captures the dynamic characteristics of user operation habits. The baseline model update under the federated learning framework, combined with differential privacy noise injection, enhances the model's generalization ability while ensuring cross-user data privacy. An intelligent mapping mechanism based on preset deviation thresholds transforms behavioral deviation into probability values, providing interpretable quantitative input for risk scoring. Adapting to cloud phone resource constraints ultimately reduces the false negative rate of spoofing attacks.
[0107] In some instances, risk level scores are calculated based on the probability of behavioral deviation and threat intelligence weights, including:
[0108] The first rating weight coefficient is determined based on the probability of user behavior deviation;
[0109] The second scoring weight coefficient is determined based on real-time risk indicators from external threat intelligence sources.
[0110] An initial risk score is generated by multiplying the first score weight coefficient by the probability of user behavior deviation, and then adding the product of the second score weight coefficient and the threat intelligence weight.
[0111] The initial risk score is normalized based on the preset risk mapping rule to determine the risk level score. The preset risk mapping rule is the correspondence between the score range and the risk level.
[0112] For example, based on the probability of user behavior deviation (a quantitative value representing the likelihood of abnormal current operation, ranging from [0,1]), a first scoring weight coefficient (denoted as k1) is determined. This coefficient is preset to a fixed value (e.g., 0.7) according to the cloud phone security policy, reflecting the dominant role of behavioral analysis in risk assessment. Simultaneously, based on real-time risk indicators (e.g., attack frequency or geographical threat level per unit time) from external threat intelligence sources (e.g., high-risk IP database, malicious behavior feature database), these indicators are quantified into threat intelligence weights (range [0,1]) through linear normalization, and a second scoring weight coefficient (denoted as k2) is determined accordingly. k2 is preset to 0.3, reflecting an auxiliary weight of the external threat situation, and satisfies the constraint k1+k2=1. This step ensures the integrity of the risk assessment dimensions by separating the weight allocation mechanism between internal behavioral anomalies and external environmental threats.
[0113] An initial risk score is generated through a product-plus-addition operation: First, the product of the first score weight coefficient k1 and the user behavior deviation probability P (k1×P) is calculated. Then, the product of the second score weight coefficient k2 and the threat intelligence weight T (k2×T) is added to obtain the initial risk score S0 = k1×P + k2×T. Here, the user behavior deviation probability P is directly output by the temporal feature extraction network (e.g., P = 0.8 when cosine similarity < 0.5 is detected), and the threat intelligence weight T is generated by parsing the threat level mapping of external intelligence sources (e.g., T = 0.9 when the current IP is detected to be in a high-risk area). This superposition mechanism integrates the operational habit anomaly degree with the real-time threat environment into a single score value, theoretically ranging from [0,1].
[0114] The initial risk score S0 is normalized based on a preset risk mapping rule: when S0 < 0, it is forcibly set to 0; when S0 > 1, it is forcibly set to 1, ensuring that the score value effectively falls within the [0,1] interval. Subsequently, a piecewise linear function maps the normalized score to a standard risk level interval: if S0 ≥ 0.8, it is mapped to a low risk level (score interval [0.8, 1.0]); if 0.5 ≤ S0 < 0.8, it is mapped to a medium risk level (score interval [0.5, 0.8]); if S0 < 0.5, it is mapped to a high risk level (score interval [0, 0.5]). This mapping rule is preset in the strategy library in the form of numerical intervals, achieving a deterministic conversion from score to risk level.
[0115] Based on the calibrated normalized score and preset risk mapping rules, the final risk level score is output. This score retains the original numerical value (e.g., S0 = 0.75, output is 0.75) while also associating it with the corresponding risk level label (e.g., 0.75 belongs to the medium risk level). The scoring result is stored as a structured data object, containing the original score value, normalized score value, risk level identifier, and timestamp, for subsequent use by the authentication strategy selection module. This process ensures the traceability of the scoring output and avoids misjudgment of the level due to floating-point operation errors.
[0116] In summary, this application's embodiments achieve dynamic fusion of user operation habit anomalies and environmental threat situations by separating the weight allocation of internal behavioral deviation probability and external threat intelligence. After generating an initial score through product superposition, it undergoes forced normalization and preset interval mapping rules to eliminate the influence of extreme values and clearly classify low / medium / high risk levels (thresholds of 0.5 and 0.8). Finally, a structured score with both numerical accuracy and level identification is output, providing a quantitative basis for the dynamic selection of multi-factor authentication strategies. This method overcomes the limitations of traditional static threshold determination, achieving improved risk assessment accuracy with a computational latency of less than 5ms in resource-constrained cloud phone environments, reducing the false negative rate of high-risk operations, and adapting to different security strategy scenarios through a configurable weight coefficient mechanism.
[0117] In some instances, a preset multi-factor authentication strategy is selected based on risk level scoring, including:
[0118] Based on the risk level score and the first preset threshold, the first risk level range is determined;
[0119] The second risk level range is determined based on the risk level score and the second preset threshold.
[0120] When the risk level score is in the first risk level range, the target authentication strategy is determined to be the first type of authentication strategy, which combines static passwords and dynamic tokens.
[0121] When the risk level score is in the second risk level range, the target authentication strategy is determined to be the second type of authentication strategy that adds device fingerprint verification.
[0122] When the risk level score is less than the second preset threshold, the target authentication strategy is determined to be the third type of authentication strategy that combines voiceprint recognition and geofence verification.
[0123] For example, risk level intervals are dynamically divided based on the relationship between a risk level score (denoted as S, ranging from [0,1]) and a preset threshold. Specifically: the first risk level interval is defined as S ≥ the first preset threshold (the first preset threshold is set to 0.8), representing a low-risk scenario; the second risk level interval is defined as the second preset threshold ≤ S < the first preset threshold (the second preset threshold is set to 0.5, and the first preset threshold remains at 0.8), representing a medium-risk scenario; and a high-risk scenario is defined as S < the second preset threshold (i.e., S < 0.5). The threshold settings are based on historical attack data analysis: when the score ≥ 0.8, the device hijacking probability is less than 1%, and user operations can be trusted; when the score < 0.5, the hijacking probability exceeds 20%, requiring the highest level of defense. The thresholds are fixed in the strategy library to ensure the determinism of the division logic.
[0124] When the risk level score falls within the first risk level range (S≥0.8), the target authentication strategy is determined to be a first-type authentication strategy, namely a two-factor authentication (1FA) combination of static password and dynamic token. Specific execution includes: Static password verification: the user inputs a preset password string, and the server compares the hash value; Dynamic token verification: a token is generated using a time-window-based TOTP algorithm (60-second time window), and the server verifies the timeliness after the client submits the token. This strategy, while ensuring basic security, avoids introducing additional verification steps (such as biometrics) and maintains smooth user operation. Simultaneously, the session expiration time is extended from the default 5 minutes to 2 hours, reducing interference from repeated authentication in low-risk scenarios.
[0125] When the risk level score falls into the second risk level range (0.5 ≤ S < 0.8), the target authentication strategy is determined to be the second type of authentication strategy, which is to add device fingerprint verification (upgraded to 2FA) on top of the first type of authentication strategy. Specifically, this includes: static password and dynamic token verification: inheriting the verification logic of the first type of strategy; device fingerprint binding verification: generating a device fingerprint (512 bits long) in real time based on a preset anti-collision algorithm (such as TLSH). The server compares the fingerprint with historical registration records; if the deviation exceeds the Hamming distance threshold (preset 120 bits), authentication is rejected. This strategy defends against device forgery or hijacking attacks through device uniqueness verification, balancing security and user experience.
[0126] When the risk level score is less than the second preset threshold (S<0.5), the target authentication strategy is determined to be the third type of authentication strategy, which is to add voiceprint recognition and geofence verification (upgraded to 3FA) on the basis of the second type of authentication strategy. Specifically, the execution includes: static password, dynamic token, and device fingerprint verification: inheriting the verification logic of the second type of strategy; voiceprint recognition: collecting user voice segments through a microphone, extracting Mel-frequency cepstral coefficients (MFCC) features, and comparing the similarity with pre-stored voiceprint templates (threshold ≥0.85); geofence verification: obtaining the device's real-time GPS coordinates to verify whether it is within a preset security fence (e.g., a company geofence radius of 500 meters). This strategy uses strong dual verification of biometrics and physical location to deal with high-risk impersonation or cross-regional attacks. Simultaneously, it forcibly triggers biometric authentication (e.g., 3D structured light facial recognition) to ensure the authenticity of the operator's identity.
[0127] To mitigate user experience degradation caused by sudden changes in authentication policies, a tiered response rule is implemented. Upon the first detection of an anomaly (e.g., a single behavioral deviation probability P > 0.7), only the session expiration time is extended (from 5 minutes to 2 hours), without escalating authentication factors. After three consecutive anomaly detections (e.g., three consecutive P > 0.7), a forced upgrade to biometric authentication (e.g., facial recognition) is initiated, and a security log is generated. This mechanism gradually increases authentication strength, preventing frequent authentication interruptions due to brief misjudgments while ensuring proactive defense against persistent threats.
[0128] In summary, this application's embodiments achieve optimized resource adaptation through dynamic risk zone segmentation and precise matching of multi-factor strategies. In low-risk scenarios, lightweight 1FA authentication (password + token only) is used to reduce the computational load on cloud phones. In high-risk scenarios, 3FA is enabled for enhanced defense, but real-time strategy computation overhead is avoided through a pre-built strategy library. Security resilience is improved by automatically switching authentication strength based on quantitative scoring, increasing the cost of spoofing attacks in high-risk scenarios by more than three times (requiring simultaneous forgery of password, device fingerprint, voiceprint, and location). User experience is balanced; the first anomaly only extends the session time, and only consecutive anomalies force an authentication upgrade, reducing user interference in misjudgment scenarios by 60%. Furthermore, seamless verification such as geofencing reduces active interaction. Ultimately, a dynamic balance between defense strength, resource consumption, and operational smoothness is achieved while ensuring the security of cloud phone authentication.
[0129] In some instances, dynamic tokens bound to the physical characteristics of the device are generated based on a preset multi-factor authentication strategy, including:
[0130] Based on the target type of the preset multi-factor authentication strategy, determine the dynamic token algorithm type;
[0131] The device battery temperature value based on sensor data is used to generate a first salt value through a first preset scaling transformation.
[0132] The second salt value is generated based on the screen brightness value from sensor data through an 8-bit unsigned integer conversion.
[0133] An initial token sequence is generated based on the true random number seed output by the quantum random number generator;
[0134] Based on the first salt value and the second salt value, a hash binding operation is performed on the initial token sequence to obtain the hash binding operation result;
[0135] The current operation sequence identifier is determined based on the globally monotonically increasing sequence number rule;
[0136] A dynamic token is generated based on the hash binding operation result and the current operation sequence identifier to bind the physical characteristics of the device.
[0137] For example, when the risk level score is in the low-risk range (score ≥ 0.8), a time-based dynamic password algorithm is selected as the basis for token generation, with a fixed time window of 60 seconds. When the risk level score drops to the medium-high risk range (score < 0.8), the algorithm switches to an event-based dynamic password algorithm and overlays a real-time behavioral feature hash value, shortening the token validity period to a single request. The algorithm switching trigger conditions strictly follow preset anomaly threshold rules. For example, when an IP address mutation rate greater than 800 kilometers per hour or a click frequency exceeding 15 times per second is detected, the event-based algorithm is forcibly activated to deal with potential attacks. This process achieves precise matching between algorithm type and risk level through a preset strategy library, avoiding delays caused by manual configuration.
[0138] The system reads raw temperature sensor data in real time (accuracy ±0.1℃) and maps the temperature value to a salinity space through a first preset scaling transformation. Specifically, the raw temperature value is multiplied by a preset scaling factor (preferably between 800 and 1200), the product is then modulo-operated to generate a 256-bit integer, and finally converted into a hexadecimal string as the first salinity value. This scaling mechanism amplifies minute temperature variations non-linearly, ensuring that the same device generates distinctly different salinity values under different temperature conditions.
[0139] The analog signal (range 0 to 255) output from the screen brightness sensor is captured and quantized into an integer using an 8-bit unsigned integer conversion rule. Specifically, the lower 8 bits of the binary representation of the brightness value are taken directly as the raw byte of the second salt value; if the brightness value exceeds 255, a modulo 256 operation is performed to ensure that the output value is always within the range of an 8-bit unsigned integer. This design utilizes the instantaneous physical state of screen brightness to provide a dynamically changing auxiliary salt value for token binding, adapting to the brightness adjustment characteristics of mobile devices and enhancing the randomness of the token through its dynamic changes.
[0140] A physical entropy source based on quantum polarization states is used to generate a truly random number sequence at a rate of 1 gigabits per second. The entropy value is verified to be no less than 0.98 by the national information security standard NIST SP800-90B. This sequence serves as the seed source for the token generation algorithm: when using a time-based dynamic password algorithm, the counter is the integer quotient of the current UNIX timestamp divided by the time window; when using an event-based dynamic password algorithm, the first 8 bytes of the behavioral feature hash value are used as the dynamic counter. Finally, an initial token sequence is generated using a hash message authentication code algorithm.
[0141] The initial token sequence is concatenated with the first and second salt values in a preset order to form a byte stream, which is then input into the secure hash algorithm SHA3-384 for a one-way hash operation. The concatenation rule uses a structured format of first salt value + initial token sequence + second salt value to ensure that the device's physical status data completely encapsulates the core token. The 384-bit binary string output by the hash operation is the hash binding result, which is strongly bound to the device's battery temperature and screen brightness.
[0142] The server maintains a 64-bit global sequence number counter, which is atomically incremented with each authentication request. The client obtains the current counter value as an operation sequence identifier, allowing for continuity verification within a tolerance window of ±3. After the sequence number is generated, it is combined with the server's timestamp and device fingerprint to generate a globally unique identifier, and transmitted encrypted using the national cryptographic algorithm SM4 to prevent tampering.
[0143] The hash binding operation result and the operation sequence identifier are combined into a string in the format of sequence identifier:hash value, and then converted into a transmittable token text through Base32 encoding. The token payload includes an algorithm type identifier (0 indicates a time-based dynamic password algorithm, and 1 indicates an event-based dynamic password algorithm), a serial number, and a bound hash value. The total length is fixed at 512 bits to adapt to the network transmission limitations of cloud phones.
[0144] In summary, this application's embodiments achieve real-time matching of authentication strength and threat level by dynamically switching the risk scoring algorithm type (time-based dynamic password algorithm / event-based dynamic password algorithm); the device battery temperature is non-linearly scaled to generate a first salt value, and the screen brightness is converted into an 8-bit integer to generate a second salt value, deeply binding the token to the device's physical state and effectively defending against cross-device reuse attacks; quantum random numbers provide a truly random seed with an entropy value ≥0.98, combined with a globally monotonically increasing sequence number and tolerance verification mechanism, blocking replay attacks and brute-force attacks. Finally, in a resource-constrained cloud phone environment, the generation delay achieves dual protection of token unpredictability and device uniqueness.
[0145] In some instances, user authentication is performed based on dynamic tokens to determine the user's identity verification result, including:
[0146] The sequence continuity verification result is determined based on the current operation sequence identifier in the dynamic token and the server-side globally monotonically increasing sequence number.
[0147] The token hash reconstruction value is determined based on the device battery temperature value, screen brightness value, and quantum random number generator.
[0148] The hash consistency verification result is determined based on the hash binding operation result of the token hash reconstruction value and the dynamic token;
[0149] The user authentication result is determined by performing a logical AND operation between the sequence continuity verification result and the hash consistency verification result.
[0150] For example, the server maintains a 64-bit globally monotonically increasing sequence number counter, atomically incrementing this counter value each time an authentication request is received. The client extracts the current operation sequence identifier (i.e., the sequence number embedded when the token is generated) from the dynamic token and compares it in real time with the server's current counter value to verify whether the sequence number is within a preset tolerance window (allowing a sequence number deviation of ±3; for example, if the server records a value of 100, then identifiers between 97 and 103 are accepted). If the client's sequence number is equal to the server's counter value, or within the closed interval of the counter value minus 3 to plus 3, the sequence is considered continuous, and a sequence verification result is output. If the sequence number exceeds the tolerance window (e.g., less than the historical counter value minus 3 or greater than the current counter value plus 3), a replay attack or man-in-the-middle attack risk is identified, and a sequence verification result is output. This mechanism ensures the uniqueness and non-repeatability of each authentication request through strict temporal constraints on the operation sequence.
[0151] The system acquires raw data from the device's battery temperature sensor in real time (measurement accuracy ±0.1 degrees Celsius). A first salt value is generated through a first preset scaling transformation (the preferred scaling factor is between 800 and 1200). Specifically, the raw temperature value is multiplied by the scaling factor, modulo 256, and converted to a hexadecimal string. Simultaneously, the output value from the screen brightness sensor (range 0 to 255) is acquired, and the lower 8 bits of its binary representation are extracted as the second salt value, directly participating in the calculation as an unsigned integer byte. A quantum random number generator is invoked, and a true random seed (entropy value not less than 0.98 as tested by the NIST SP800-90B standard) is reproduced based on the physical entropy source of the photonic quantum polarization state during token generation. Based on the algorithm type identifier bit recorded in the dynamic token (0 indicates a time-based dynamic password algorithm, 1 indicates an event-based dynamic password algorithm), and combined with the original counter parameters (time-based dynamic password algorithms use the integer quotient of the current UNIX timestamp divided by a 60-second time window, while event-based dynamic password algorithms use the first 8 bytes of the behavioral feature hash value), the initial token sequence is regenerated. This process completely replicates the salt value processing logic during token generation, ensuring that the reconstruction conditions are consistent with the original binding.
[0152] The generated first salt value, initial token sequence, and second salt value are concatenated into a byte stream in a fixed order (first salt value + initial token sequence + second salt value). This stream is then input into the secure hash algorithm SHA3-384 for a one-way hash operation, outputting a 384-bit binary string as the token hash reconstruction value. The original hash binding operation result (i.e., the 384-bit hash value calculated during token generation) is parsed from the dynamic token payload and compared bit by bit with the reconstruction value. If they match perfectly, it indicates that the device's physical state (battery temperature, screen brightness) has not been tampered with and the token generation environment is legitimate, and a hash verification pass result is output. If any bit differs, the device is determined to be in an abnormal environment (such as simulator forgery or sensor data deception), and a hash verification failure result is output. This verification process ensures that the token cannot be used outside its original device environment through the strong binding relationship between the salt value and the device's physical state.
[0153] A logical AND operation is performed on the sequence continuity verification result and the hash consistency verification result. Only when both the sequence verification result and the hash verification result pass (the reconstructed token hash value is completely consistent with the original value) is the user authentication successful result output, authorizing access to cloud phone resources. If either verification result fails, a user authentication failure result is output, triggering the virtual environment isolation and attack chain forensics process. This dual verification mechanism constructs a closed-loop judgment from two orthogonal dimensions: the legality of the operation sequence (defending against replay attacks) and the authenticity of the device's physical state (defending against environment forgery), ensuring that the identity is considered legitimate only when both conditions are met simultaneously.
[0154] In summary, this application's embodiments effectively block replay attacks and man-in-the-middle attacks through tolerance verification (±3 windows) using a globally monotonically increasing sequence number. It utilizes dynamic salt values generated by scaling device battery temperature (800–1200 times) and converting screen brightness to 8-bit integers to deeply bind the token hash reconstruction value to the device's physical state, completely eliminating the risks of cross-device reuse and emulator forgery. A quantum random number generator provides a truly random seed with an entropy value no less than 0.98, ensuring the unpredictability of the token generation process. The logic and operations of the dual verification results ensure the completeness of identity determination; failure in any step triggers security isolation. Ultimately, in cloud phone resource-constrained environments, it achieves a reduction in authentication error rate with low verification latency, increasing the breakthrough cost of spoofing attacks in high-risk scenarios by more than three times. Simultaneously, through lightweight hash operations (SHA3-384) and sequence number comparison logic, it adapts to the computing power constraints of edge nodes, providing a feasible closed-loop authentication support for a dynamic zero-trust architecture.
[0155] In some instances, it also includes:
[0156] When an abnormal operation is detected, a virtual environment isolation operation is triggered based on the risk level score.
[0157] For example, when the risk level score is less than a second preset threshold (i.e., a high-risk scenario, score < 0.5), a virtual environment isolation operation is triggered based on the risk level score. Specifically, this includes: creating a resource-constrained micro virtual machine using Kata Containers and configuring the CPU. Cgroups are limited to 0.5 cores, a memory limit of 512MB, and a disk I / O rate ≤10MB / s to build a lightweight containerized isolation environment. An eBPF program is loaded into the kernel flow control layer (TC) to implement a zero-trust micro-segmentation strategy, discarding all SYN packets from non-whitelisted IPs (the default sandbox gateway IP is 10.0.0.1:443), allowing only encrypted communication (port 443) between the isolation environment and the sandbox gateway. Forged data is injected into the isolation environment: a forged address book containing 50 virtual contacts is generated using a generative adversarial network (GAN), and a fake document with a tracking watermark (MD5 prefix 1a3f) is implanted to trace the attacker's operation path through a watermark association mechanism. Bidirectional traffic mirroring is automatically triggered, redirecting attack traffic to the honeypot server (masquerading as 192.168.1.100), while simultaneously recording attack chain data (including attacker IP, operation sequence SHA256 hash value, and isolation environment forensic image hash value), generating STIX-compliant data. 2.1 Standardized attack-pattern threat reporting enables a closed-loop process for attack trapping and forensics.
[0158] Please see Figure 2 The diagram below illustrates the structure of an authentication device for a cloud phone, as provided in this embodiment of the application, including:
[0159] The behavior data acquisition unit 21 is used to acquire user behavior data of the cloud phone hardware layer, including user interaction characteristics, device fingerprint, network environment data and sensor data.
[0160] Deviation probability generation unit 22, based on user behavior data, models user behavior through a time-series feature extraction network to generate user behavior deviation probability;
[0161] Risk level scoring unit 23 calculates the risk level score based on the probability of behavioral deviation and threat intelligence weights;
[0162] The authentication strategy selection unit 24 selects a preset multi-factor authentication strategy based on the risk level score.
[0163] The dynamic token generation unit 25 generates a dynamic token bound to the physical characteristics of the device based on a preset multi-factor authentication strategy.
[0164] The identity verification unit 26 performs an authentication operation on the user's identity based on the dynamic token and determines the user's identity verification result.
[0165] Please see Figure 3 This application also provides an electronic device 300, including a memory 310, a processor 320, and a computer program 311 stored in the memory 310 and executable on the processor. When the processor 320 executes the computer program 311, it implements the steps of the cloud phone authentication method.
[0166] Since the electronic device described in this embodiment is the device used to implement the authentication device of a cloud phone in the embodiments of this application, those skilled in the art can understand the specific implementation method and various variations of the electronic device in this embodiment based on the method described in the embodiments of this application. Therefore, how the electronic device implements the method in the embodiments of this application will not be described in detail here. Any device used by those skilled in the art to implement the method in the embodiments of this application is within the scope of protection of this application.
[0167] In practice, when the computer program 311 is executed by the processor, it can implement any of the embodiments corresponding to the first aspect.
[0168] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.
[0169] Those skilled in the art will understand that embodiments of this application can provide methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-readable storage media containing computer-readable program code.
[0170] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0171] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0172] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0173] This application also provides a computer program product, which includes computer software instructions that, when executed on a processing device, cause the processing device to perform... Figure 1 The flowchart of an authentication method for a cloud phone in the corresponding embodiment.
[0174] A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to the embodiments of this application is generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another. For example, computer instructions may be transferred from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless means. The computer-readable storage medium may be any usable medium that a computer can store or a data storage device such as a server or data center that integrates one or more usable media. The usable medium may be a magnetic medium, an optical medium, or a semiconductor medium, etc.
[0175] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.
[0176] In the several embodiments provided in this application, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; multiple units or components may be combined or integrated into another system, or some features may be omitted or not performed. Furthermore, the mutual couplings or direct couplings or communication connections shown or discussed may be indirect couplings or communication connections through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.
[0177] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.
[0178] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated units described above can be implemented in the form of hardware and / or software functional units.
[0179] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device to execute all or part of the steps of the methods in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory, magnetic disks, or optical disks.
[0180] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.
[0181] Although preferred embodiments have been described in this specification, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications that fall outside the scope of this specification.
[0182] Obviously, those skilled in the art can make various modifications to this specification without departing from its spirit and scope. Therefore, this specification also intends to include any modifications that fall within the scope of the claims and their equivalents.
Claims
1. A method for authenticating a cloud phone, the method comprising: include: Acquire user behavior data at the cloud phone hardware layer, wherein the user behavior data includes user interaction characteristics, device fingerprint, network environment data, and sensor data; The acquisition of user behavior data at the cloud phone hardware layer includes: Based on probes deployed in the Hypervisor layer, raw signal data from physical sensor registers is read; Based on the gyroscope and accelerometer sample values in the original signal data, a three-dimensional motion trajectory model of the sensor data is generated through a Kalman filter. The device fingerprint is generated by processing the device's unique identifier set based on a preset anti-collision algorithm, wherein the device's unique identifier set includes IMEI, MAC address, and SIM card ICCID. Based on the packet header information captured by the network interface, the network environment data is generated by parsing, wherein the network environment data includes the IP geographical location change rate and the base station handover frequency; The user interaction features are generated by extracting click coordinate distribution and swipe trajectory patterns based on the user input event stream. The three-dimensional motion trajectory model of the sensor data, the device fingerprint, the network environment data, and the user interaction features are integrated into user behavior data. Based on the user behavior data, user behavior is modeled through a time-series feature extraction network to generate user behavior deviation probability. Based on the aforementioned behavioral deviation probability and threat intelligence weight, a risk level score is calculated; Based on the risk level score, a preset multi-factor authentication strategy is selected; Based on the preset multi-factor authentication strategy, a dynamic token bound to the physical characteristics of the device is generated, including: Based on the target type of the preset multi-factor authentication strategy, determine the dynamic token algorithm type; The device battery temperature value based on sensor data is used to generate a first salt value through a first preset scaling transformation. The second salt value is generated based on the screen brightness value from sensor data through an 8-bit unsigned integer conversion. An initial token sequence is generated based on the true random number seed output by the quantum random number generator; Based on the first salt value and the second salt value, a hash binding operation is performed on the initial token sequence to obtain the hash binding operation result; The current operation sequence identifier is determined based on the globally monotonically increasing sequence number rule; Based on the hash binding operation result and the current operation sequence identifier, a dynamic token is generated that binds the physical characteristics of the device. Based on the dynamic token, the user's identity is authenticated to determine the user's authentication result, including: Based on the current operation sequence identifier in the dynamic token and the server-side globally monotonically increasing sequence number, the sequence continuity verification result is determined; Based on the device battery temperature value, the screen brightness value, and the quantum random number generator, determine the token hash reconstruction value; Based on the hash binding operation result of the token hash reconstruction value and the dynamic token, the hash consistency verification result is determined; The user authentication result is determined by a logical AND operation between the sequence continuity verification result and the hash consistency verification result.
2. The method according to claim 1, characterized in that, The step of modeling user behavior using a time-series feature extraction network based on the user behavior data and generating user behavior deviation probabilities includes: Based on the click coordinate distribution and sliding trajectory pattern in the user interaction features, an operation event sequence is generated; Based on the sequence of operation events, a behavioral feature vector is determined by a preset temporal feature extraction network. Based on the behavioral feature vector and the preset user behavior baseline model, the behavioral deviation is determined by cosine similarity calculation. Based on the mapping relationship between the behavioral deviation degree and the preset deviation threshold, the probability of user behavior deviation is determined.
3. The method according to claim 1, characterized in that, The risk level score is calculated based on the probability of behavioral deviation and the threat intelligence weight, including: The first rating weight coefficient is determined based on the user behavior deviation probability. The second scoring weight coefficient is determined based on real-time risk indicators from external threat intelligence sources. An initial risk score is generated by superimposing the product of the first scoring weight coefficient and the user behavior deviation probability, and then superimposing the product of the second scoring weight coefficient and the threat intelligence weight. The initial risk score is normalized based on a preset risk mapping rule to determine the risk level score, wherein the preset risk mapping rule is the correspondence between the score range and the risk level.
4. The method according to claim 1, characterized in that, The step of selecting a preset multi-factor authentication strategy based on the risk level score includes: Based on the risk level score and the first preset threshold, a first risk level range is determined; Based on the risk level score and the second preset threshold, a second risk level range is determined; When the risk level score is within the first risk level range, the target authentication strategy is determined to be the first type of authentication strategy that combines static password and dynamic token. When the risk level score is within the second risk level range, the target authentication strategy is determined to be the second type of authentication strategy that adds device fingerprint verification. When the risk level score is less than the second preset threshold, the target authentication strategy is determined to be the third type of authentication strategy that combines voiceprint recognition and geofence verification.
5. An authentication device for a cloud phone, used to implement the method according to any one of claims 1 to 4, characterized in that, include: The behavior data acquisition unit is used to acquire user behavior data of the cloud phone hardware layer, wherein the user behavior data includes user interaction features, device fingerprint, network environment data and sensor data. The deviation probability generation unit generates user behavior deviation probabilities based on the user behavior data by modeling user behavior through a time-series feature extraction network. The risk level scoring unit calculates a risk level score based on the probability of behavioral deviation and the threat intelligence weight. The authentication strategy selection unit selects a preset multi-factor authentication strategy based on the risk level score. The dynamic token generation unit generates a dynamic token bound to the physical characteristics of the device based on the preset multi-factor authentication strategy. The identity verification unit performs an authentication operation on the user's identity based on the dynamic token and determines the user's identity verification result.
6. An electronic device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program stored in the memory to implement the steps of the authentication method for a cloud phone as described in any one of claims 1 to 4.
7. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the authentication method for cloud phones as described in any one of claims 1 to 4.
Citation Information
Patent Citations
Account-free user unification method and system of construction and management system based on multimode authentication
CN119357939A
